CN104700373B - A kind of multiple dimensioned gaussian filtering device and its building method for image analoging signal - Google Patents
A kind of multiple dimensioned gaussian filtering device and its building method for image analoging signal Download PDFInfo
- Publication number
- CN104700373B CN104700373B CN201510138175.3A CN201510138175A CN104700373B CN 104700373 B CN104700373 B CN 104700373B CN 201510138175 A CN201510138175 A CN 201510138175A CN 104700373 B CN104700373 B CN 104700373B
- Authority
- CN
- China
- Prior art keywords
- gaussian filtering
- pixel
- filter layer
- sampling filter
- signal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Image Processing (AREA)
Abstract
The invention discloses a kind of multiple dimensioned gaussian filtering device for image analoging signal, described device is used to carry out multiple dimensioned gaussian filtering to pending image analoging signal so as to obtain filtered image analoging signal, wherein:Described device includes multiple gaussian filtering units;The negative resistance and common positive resistance that each gaussian filtering unit is made up of multiple active circuits connect and compose according to specific internetwork connection mode.The invention also discloses a kind of building method of the multiple dimensioned gaussian filtering device for image analoging signal.Compared with prior art, gaussian filtering process is carried out using the device of the present invention, not only substantially increases processing speed, and reduce processing energy consumption.
Description
Technical field
The present invention relates to audio- visual technique field, in particulars relate to a kind of multiple dimensioned gaussian filtering for image analoging signal
Device and its building method.
Background technology
Especially it is applied in some pattern treatment procedures in the pattern treatment procedure of feature extraction algorithm, it usually needs right
Graph data carry out multi-sampling rate under and it is multiple dimensioned under gaussian filtering process.It is traditional in order to realize above-mentioned filtering process
Graphic system first has to construct corresponding gaussian pyramid in systems based on software algorithm more when carrying out graphics process
Scale filter module, the image analoging signal collected is then converted into data signal, finally utilizes the more chis of gaussian pyramid
Spend filtration module and corresponding calculating processing is carried out to data signal.
Because Gaussian filter algorithm possesses larger amount of calculation, therefore in above process, the more chis of gaussian pyramid are constructed
Spend filtration module and carry out calculating processing using gaussian pyramid multi-scale filtering module and all cause huge calculating consumption.
Such as utilizing scale invariant feature conversion (Scale-invariant feature transform, SIFT) algorithm process figure
As during, the process of structure gaussian pyramid multi-scale filtering module occupies about the 85.63% of whole algorithm.
The disposal ability of graphic system is mainly reflected in the real-time and processing energy consumption of data processing, is treated
Calculating consumption in journey certainly will cause corresponding energy consumption and time loss.With surging for currently pending graph data amount,
Energy consumption caused by gaussian filtering process and time loss are to graphic system entirety ability in pattern treatment procedure
Influence more and more prominent.
Therefore, in order to improve the disposal ability of graphic system, it is necessary to one kind relative to prior art energy consumption less part
Manage the faster gaussian filtering device of speed.
The content of the invention
The problem of existing for the gaussian filtering process for being directed to picture signal in the prior art, the invention provides a kind of pin
To the multiple dimensioned gaussian filtering device of image analoging signal, described device is used for multiple dimensioned to the progress of pending image analoging signal
Gaussian filtering so as to obtain filtered image analoging signal, wherein:
Described device includes multiple gaussian filtering units;
Negative resistance that each gaussian filtering unit is made up of multiple active circuits and common positive resistance are according to specific
Internetwork connection mode connects and composes.
In one embodiment, the pending image analoging signal includes multiple pixel analog signals, the Gauss filter
The gaussian filtering process that ripple unit is configured to carry out under specific bandwidth a pixel analog signal is filtered to obtain
Pixel analog signal.
In one embodiment, described device includes sampling filter layer, and the sampling filter layer is used to be based on particular sample rate
Obtain corresponding multiple pixel analog signals and the multiple pixel to getting in the pending image analoging signal
Point analog signal carries out gaussian filtering process.
In one embodiment, described device includes multiple sampling filter layers, each corresponding spy of the sampling filter layer
Fixed sample rate, the different sampling filter layers correspond to different sample rates.
In one embodiment, described device includes a sampling filter layer and multiple sampling extraction modules, the sampling
Extraction module is used to filter out specific pixel from the filtered pixel analog signal of sampling filter layer output
Corresponding analog signal is so as to obtaining the filtered image analoging signal under corresponding sample rate.
In one embodiment, the sampling filter layer includes multiple dimensioning filtration modules, the dimensioning filtration module
Multiple pixel analog signals for being got with specific bandwidth to the sampling filter layer carry out gaussian filtering process.
In one embodiment, each dimensioning filtration module includes multiple gaussian filterings with same band
Unit, got with sampling filter layer described in each gaussian filtering unit alignment processing in some scale filtration module multiple
A pixel analog signal in pixel analog signal, in any one dimensioning filtration module of the sampling filter layer
In, the gaussian filtering unit corresponds with each pixel analog signal that the sampling filter layer is got.
In one embodiment, the sampling filter layer includes multiple pixel filter modules, every in the sampling filter layer
Individual pixel filter module corresponds with each pixel analog signal that the sampling filter layer is got.
In one embodiment, the pixel filter module includes multiple gaussian filtering units, same pixel filter mould
Different gaussian filtering units in block have a different bandwidth, and all pixels filtration module in same sampling filter layer has phase
Same structure.
It is described the invention also provides a kind of building method of the multiple dimensioned gaussian filtering device for image analoging signal
Method comprises the steps of:
Analyzing and processing demand step, according to the process demand of pending image determine filtering process sample rate requirement and
Bandwidth demand;
Determining device configuration steps, according to each Gauss in the sample rate requirement and the bandwidth demand determining device
The network connection mode of filter processing unit;
Gaussian filtering process cell parameters step is determined, dress is determined according to the sample rate requirement and the bandwidth demand
Put the quantity of middle gaussian filtering process unit and the particular procedure bandwidth of each gaussian filtering process unit;
Gaussian filtering process unit step is constructed, connects according to specific internetwork connection mode and is made up of multiple active circuits
Negative resistance and common positive resistance so as to forming the gaussian filtering process unit with the particular procedure bandwidth;
Constructing apparatus step, it is described so as to form that the gaussian filtering process unit is connected based on the network connection mode
Device.
Compared with prior art, gaussian filtering process is carried out using the device of the present invention, not only substantially increases processing speed
Degree, and reduce processing energy consumption.
The further feature or advantage of the present invention will illustrate in the following description.Also, the present invention Partial Feature or
Advantage will be become apparent by specification, or be appreciated that by implementing the present invention.The purpose of the present invention and part
Advantage can be realized or obtained by specifically noted step in specification, claims and accompanying drawing.
Brief description of the drawings
Accompanying drawing is used for providing a further understanding of the present invention, and a part for constitution instruction, the reality with the present invention
Apply example to be provided commonly for explaining the present invention, be not construed as limiting the invention.In the accompanying drawings:
Fig. 1 is apparatus according to an embodiment of the present invention structure diagram;
Fig. 2 is apparatus according to an embodiment of the present invention partial circuit sketch;
Fig. 3 is apparatus according to an embodiment of the present invention partial circuit sketch;
Fig. 4 is apparatus according to an embodiment of the present invention partial circuit sketch;
Fig. 5 is method flow diagram according to an embodiment of the invention;
Fig. 6 is apparatus according to an embodiment of the present invention partial circuit line schematic diagram;
Fig. 7 is apparatus according to an embodiment of the present invention partial circuit line schematic diagram;
Fig. 8 is apparatus according to an embodiment of the present invention structure diagram.
Embodiment
Embodiments of the present invention are described in detail below with reference to drawings and Examples, whereby implementation personnel of the invention
Can fully understand how application technology means solve technical problem to the present invention, and reach the implementation process of technique effect and according to
The present invention is embodied according to above-mentioned implementation process.If it should be noted that do not form conflict, each embodiment in the present invention
And each feature in each embodiment can be combined with each other, the technical scheme formed protection scope of the present invention it
It is interior.
Traditional gaussian filtering module is to be built in graphic system based on software algorithm.Not only its building process meeting
A large amount of amounts of calculation are brought, and when it is filtered processing, with the increase of graph data amount, the amount of calculation of its filtering process
Also increase considerably.In order to improve the disposal ability of graphic system, the invention discloses a kind of gaussian filtering device and its structure
Make method.
The present invention constructs the gaussian filtering device of particular characteristic parameter first.So can be direct when carrying out graphics process
Using the gaussian filtering device constructed, so as to eliminate the construction step of gaussian filtering module in conventional method.
In the prior art, the first step of image procossing is typically that the optical signal of pending image first is converted into simulation letter
Number.Because prior art is to carry out gaussian filtering process using software algorithm, therefore needed in existing image processing system
Image analoging signal (voltage amplitude signal) is changed into data signal.When the data signal to image carries out gaussian filtering,
First have to build the process of gaussian filtering module not using software algorithm structure gaussian filtering module for specific processing requirement
It is evitable to generate substantial amounts of calculating.In order to reduce amount of calculation, the present invention is no to carry out gaussian filtering using software algorithm
Processing.But active pull-up structure gaussian filtering network is used, using gaussian filtering network directly to image analoging signal (voltage
Range signal is handled), so as to realize gaussian filtering.
As shown in figure 1, light stream detector 101 includes multiple pixel cells 102,103,104,105 ....Pixel cell is light
According to sensor, it experiences light stream signal, and light stream signal is converted into voltage amplitude signal.It is each in light stream detector 101
Pixel cell is directed to a pixel in pending image, and it changes the light stream signal in each pixel in pending image
For pixel analog signal (voltage amplitude signal).All pixels point analog signal collectively forms image analoging signal (signal
150).Device 100 is directly handled signal 150.
For the carry out gaussian filtering process to analog signal, present invention employs construct gaussian filtering based on hardware mode
The method of network realizes gaussian filtering.In the present embodiment, negative resistance is realized using active circuit, using negative resistance and commonly
Resistance realizes Gaussian filter circuit network together.Gaussian circuit network realizes approximate Gaussian curve using pure resistance device,
So as to realize filtering application.
The composition that the device of the present embodiment is most basic is gaussian filtering unit.Device includes multiple gaussian filtering units, often
Individual gaussian filtering unit is configured to carry out specific bandwidth to a pixel analog signal in pending image analoging signal
Under gaussian filtering process to obtain filtered pixel analog signal.By the gaussian filtering network of multiple gaussian filtering units
A catenet is connected and composed according to ad hoc fashion, it is possible to realizes the multiple dimensioned gaussian filtering under multi-sampling rate.
In the present embodiment, gaussian filtering unit is made up of two common positive resistances and a negative resistance, wherein negative resistance
Realized by active circuit.One Gaussian filter circuit network is connected from several above-mentioned elementary cells to horizontally and vertically duplication extension
Connect and form, these elementary cells are full symmetric, the equal correspondent equals of resistance value.Two positive electricity in elementary cell in above-mentioned
The proportionate relationship of resistance will determine the filtering bandwidth of a dimensional Gaussian circuit network.So, the gaussian circuit of different filter scales
The difference of network is substantially that the resistance ratio relation in its gaussian filtering unit that forms is different.The circuit of different filter scales
Network connects and composes the multi-scale filtering network under same size, the different scale filtering net of same size according to ad hoc fashion
The corresponding sampling filter layer of network.Multi-scale filtering network interconnection under various sizes of circuit scale, form whole more samplings
Rate Gaussian filter.
First from one-dimensional gaussian filtering Crosslinking Structural, as shown in Fig. 2 n-th of nodes X n passes through as Centroid
Positive resistance R1 is connected with adjacent node Xn-1 and Xn+1, is connected by negative resistance R2 with time adjacent node Xn-2 and Xn+2,
It is connected by R0 with input Un.Input is Un-1 in circuit structure, the magnitude of voltage such as Un, Un+1, is exported as nodes X n-1, Xn,
The node voltages such as Xn+1.R0, R1 are positive resistance, and R2 is negative resistance, and proportionate relationship R2/R1 is a specific negative value, R0's
Size controls the bandwidth of Gaussian curve, and the structure is symmetrical structure, can infinitely be extended to both sides.Voltage source given voltage, treats electricity
After road is stable, the magnitude of voltage of output node is the result after gaussian filtering.
2-d gaussian filterses network is generalized to, in order to ensure the circular symmetry of gaussian filtering curve in two-dimensional network, is being incited somebody to action
One dimensional network towards X-direction and Y-direction extend simultaneously while, it is also contemplated that diagonal.As shown in figure 3, Centroid P0 with
Adjacent node (P1, P2, P3, P4, P5, P6, P7 and P8) is connected by positive resistance R1, with secondary adjacent node (P9, P10, P11,
P12, P13, P14, P15 and P16) it is connected by negative resistance R2, it is connected with input U0 by resistance R0.
The electric resistance structure of the 2-d gaussian filterses network of the present embodiment is designed with several vital points:
1. the design of negative resistance, is realized using active pull-up in the design.As shown in figure 4, negative resistance is symmetrical etc. by two
Imitate resistance R2a and R2b to form, realize both-end symmetry.Both sides circuit structure is equivalent to a stream control voltage source, its pair respectively
Hold the negative value that equivalent resistance is interlaminated resistance R2.Different from the design of traditional negative resistance, the negative resistance in this structure has taken into account net
The symmetry of network, complete " hanging " effect is devised, makes that all there is negative resistance effect in terms of two ports.In the present embodiment, it is right
Equivalent resistance R2a and R2b are referred to as current-controlled voltage source (Current Controlled Voltage Source, CCVS).
2. design of network topology structure, the structure as shown in Fig. 2 or Fig. 3, in the case of no introducing negative resistance, system letter
Number curve downward trend is relevant with resistance R0, R1 size, and now circuit is a simple pure resistor element circuit.For penalty function
Curve, force it to approach Gaussian curve, introduce negative resistance (R2) compensation, by tight mathematical derivation, ensure that R2 and R1 ratio is closed
System and R2 access relation, you can function curve is approached Gaussian function curve.
3. the connection extension in two-dimensional network on diagonal, structure as shown in Figure 3, it is contemplated that network topology is two dimensional wireless
Grid network (Mesh) structure, the circular symmetry feature of function curve is taken into account, same extension has been done in diagonal, so may be used
So that the voltage downward trend on diagonal approaches X and Y-direction, the accuracy of boostfiltering.The voltage of each node and its
The adjoining several node voltage relations of surrounding are very close, and network edge can not realize that surrounding is symmetrical, therefore on edge
Node voltage curve has a certain degree of deformity.On the basis of one-dimensional gaussian filtering network, its parameter is reconfigured, filled
Divide and realize multi-scale filtering function using R0 flexibility, design interconnected relationship, the two-dimensional network for designing multilayer different scales is mutual
Connect and realize gaussian pyramid, every layer of filter network (the R0 values of i.e. each two-dimensional network are different) for including multiple different bandwidths,
The multiple dimensioned gaussian filtering task in image characteristics extraction is completed with this.
For each node in Fig. 3, each input voltage of its filter result and the two-dimensional filtering network is relevant,
The node centered on P0, it is connected by positive resistance R1 with 8 adjacent nodes (P1, P2, P3, P4, P5, P6, P7 and P8), is led to
Cross negative resistance R2 and be connected with being separated by the node (P9, P10, P11, P12, P13, P14, P15 and P16) of a node, realize voltage
Compensation, is connected by resistance R0 with input U0, wherein resistance R0 sizes control filtering bandwidth.After circuit establishes stabilization,
The output voltage of corresponding node is the result for filtering and completing, and after these voltages are read, carries out subsequent treatment, completes Gauss
Filtering algorithm.
In digital filter system in the prior art, complete filter task and need to use filtration module to scan input number
According to the result that convolutional calculation (being substantially exactly multiply-add calculating) obtains is as output.Processing method is very time-consuming, meanwhile, so
More multiply-add operations will bring larger energy consumption.Compared to prior art, in the present embodiment based on pure resistance device institute structure
Energy consumption of the gaussian filtering network founded a capital during processing data and time-consuming substantially reduce.
Based on above-mentioned principle, the invention discloses a kind of method based on hardware mode construction Gauss device.The present invention's
Building method implementation procedure is as illustrated in the flow chart of figure 5.The step of being shown in the flow chart of accompanying drawing can include such as one group
Performed in the computer system of computer executable instructions.Although showing the logical order of each step in flow charts,
In some cases, can be with the step shown or described by being performed different from order herein.
In the prior art, gaussian filtering process is typically to multiple under the multiple different sample rates of pending figure progress
The gaussian filtering of different bandwidth is so as to obtaining a whole set of filter result.Therefore, two crucial parameters of gaussian filtering process are set
Surely it is sample rate requirement and bandwidth demand, for different image or image processing requirements, it is necessary to the gaussian filtering carried out
The sample rate and bandwidth of processing are different.
As shown in figure 5, the building method for performing the gaussian filtering device of the present invention first has to perform step S510, at analysis
Reason demand step, the sample rate requirement and bandwidth demand of filtering process are determined according to the process demand of pending image.
Then can performs step S520, and determining device configuration steps are true according to sample rate requirement and bandwidth demand
Determine the network connection mode of each gaussian filtering process unit in device.
In the present embodiment, for sample rate requirement, device includes sampling filter layer, and sampling filter layer is used for based on specific
Sample rate obtains corresponding multiple pixel analog signals and multiple pixels to getting in pending image analoging signal
Analog signal carries out gaussian filtering process., it is necessary to carry out the Gauss under multiple sample rates to figure in common Gauss processing
Processing.Therefore, the device in the present embodiment includes multiple sampling filter layers, the corresponding one specific sampling of each sampling filter layer
Rate, different sampling filter layers correspond to different sample rates.
As shown in figure 1, filter 100 includes sampling filter layer 111,112,113 ....The wherein sampling of wave filtering layer 111
Rate is maximum in the sample rate of all wave filtering layers in device 100.Sampling filter layer 111,112,113 ... is according to sample rate size
It is descending to be arranged in order.In general, the maximum sample rate of filter and the ultimate resolution of pending figure are
Consistent.In the present embodiment, (figure signal input 150 is 256 × 256 pictures to the figure that pending figure is 256 × 256
Vegetarian refreshments signal), then the sample rate of sampling filter layer 111 is 256 × 256, and the sampling input 151 of sampling filter layer 111 is 256
× 256 pixel signals.I.e. sampling filter layer 111 needs the signal for 256 × 256 pixels to be handled.
Because output required for sampling filter layer 112 is sampling filter layer 111 by sampled result, therefore sampling filter
The sampling input 152 of layer 112 is the down-sampled of the output 151 of sampling filter layer 111, equally.The sampling of sampling filter layer 113 is defeated
Enter the down-sampled of 153 outputs 152 for being sampling filter layer 112.
In the present embodiment, sample rate requirement is since highest sample rate, and each sample rate rank is relative to upper one sampling
The sample rate that rate grade drops are at half, i.e., drop to 128 × 256 from 256 × 256.Assuming that the present embodiment is in such case, then that
The sample rate of sampling filter layer 112 is 128 × 256, and the sampling input 152 of sampling filter layer 112 is 128 × 256 pixels
Point signal.So the sample rate of sampling filter layer 113 is 64 × 256, and the sampling input 153 of sampling filter layer 113 is 64 × 256
Individual pixel signal.
As shown in fig. 6,610 be that the sampling filters of two adjacent sample rate ranks is corresponded in the present embodiment device with 620
A mutual corresponding part for layer input port.Black round dot represents the input signal interface of sampling filter layer in Fig. 6, each black
Color round dot accesses a pixel signal.Possess 5 × 3 pixel signal input signal interfaces on 610, input 611,612,
613rd, the 614 and 615 input signal interface for representing a row three respectively.
Interlayer interconnection needs to consider the size relationship between different layers, and the interconnected relationship between adjacent two layers is down-sampled pass
System, i.e., last layer is the down-sampled result of next layer.620 sample rate be 610 half, each input node of last layer
With next layer every a node it is corresponding connection.Then its interlacing selection input for actually inputting on 610 of the input on 620.I.e.
Possess 3 × 3 pixel signal input signal interfaces on 610, input 621,622 and 623, represent the defeated of a row three respectively
Enter signaling interface.621 are connected with 611 is connected respectively to three pixel signal input signals of identical, and 622 are connected point with 612
Three pixel signal input signals of identical are not connected to, and 623 are connected with 613 is connected respectively to three pixel letters of identical
Number input signal.
The device of the present embodiment is using for the independent sampling filter layer of each sample rate requirement structure.So do
Benefit is to ensure that the independence of data processing.Contribute to stable process signal, reduce the signal interference of different disposal process.
In order to simplify device internal structure, in another embodiment of the invention, device employs single sampling filter layer multi-sampling rate
The pattern of output realizes multi-sampling rate.In the device of another embodiment, a sampling filter layer and multiple samplings are included
Extraction module, sampling extraction module are specific for being filtered out from the filtered pixel analog signal of sampling filter layer output
Pixel corresponding to analog signal so as to obtaining the filtered image analoging signal under corresponding sample rate.
In common gaussian filtering process, same signal is also needed to carry out multi-scale filtering, i.e., multiple different
Gaussian filtering process (bandwidth demand) is carried out to pixel signal under bandwidth.For bandwidth demand, the device of the present embodiment is adopted
Sample wave filtering layer includes multiple dimensioning filtration modules, and dimensioning filtration module is used to obtain sampling filter layer with specific bandwidth
The multiple pixel analog signals arrived carry out gaussian filtering process.As shown in figure 1, sampling filter layer 111 filters comprising dimensioning
Module 121,122,123,124 ....
To shorten the quantity of interconnection line as far as possible, the device of the present embodiment is by way of similar bus in same sampling filter
Interconnected between each dimensioning filtration module in layer.Physical location between resistor network is network in layer in X direction
Extension, using the structure of similar bus interconnection, respective signal bus is correspondingly drawn from each pixel of signal acquisition terminal, by layer
The input carry of each resistor network corresponding node is in bus.As shown in fig. 7,710 with 720 be respectively the present embodiment in it is same
A part for any two dimensioning filtration module input port in sampling filter layer.710 and 720 respectively comprising corresponding 9
Nine input ports of pixel analog signal.710 is corresponding with each input port in 720 connected, is commonly connected to one
In the input of pixel analog signal.
Because each dimensioning filtration module is to the letter of the sampling filter layer all pixels point to be dealt with where it
Number handled, therefore each dimensioning filtration module of the device of the present embodiment includes multiple height with same band
This filter unit, got with each gaussian filtering unit alignment processing sampling filter layer in some scale filtration module more
A pixel analog signal in individual pixel analog signal, in any one dimensioning filtration module of sampling filter layer
In, gaussian filtering unit corresponds with each pixel analog signal that sampling filter layer is got.
As shown in figure 1, dimensioning filtration module 121 includes gaussian filtering unit 131,132,133 ....Described above
Assuming that in situation (figure ultimate resolution is 256 × 256), filtered in dimensioning filtration module 121 comprising 256 × 256 Gausses
Ripple unit.
The device of the present embodiment is using sample rate-bandwidth-pixel (sampling filter layer-dimensioning filtration module-Gauss
Filter unit) successively progressive pyramid structure.Explanation is needed exist for, the structure of device of the invention is not limited to this
Structure described in embodiment.In another embodiment of the invention, using the successively progressive of sample rate-pixel-bandwidth
Structure.I.e. in the device of another embodiment of the present invention:
Device is built with sampling filter layer, and sampling filter layer includes multiple pixel filter modules, every in sampling filter layer
Individual pixel filter module corresponds with each pixel analog signal that the sampling filter layer is got;
Pixel filter module includes multiple gaussian filtering units, the different gaussian filtering units in same pixel filter module
With different bandwidth, all pixels filtration module in same sampling filter layer has identical structure.
Fig. 8 show the part-structure of another embodiment of the present invention.As illustrated, sampling filter layer 812 includes pixel
Filtration module 824,825,826 and 827;Sampling filter layer 811 includes pixel filter module 822 and 823;Sampling filter
Layer 810 includes pixel filter module 821.Each pixel filter module (821-827) includes two gaussian filtering units (831
With 832,833 and 834,835 and 836,837 and 838,839 and 841,842 and 843,844 and 845).
Picture element signal inputs 801,802,803 and 804 and represents 4 different pixel signal inputs.Pixel filter mould
Block 821,822 and 824 is connected in picture element signal input 801, and pixel filter module 823 and 826 is connected to picture element signal input
On 803, pixel filter module 825 and 827 is coupled with picture element signal input 802 and 804.
Structure determination inside device well after can perform step S530, determine gaussian filtering process cell parameters walk
Suddenly, according to the quantity of gaussian filtering process unit and each gaussian filtering in sample rate requirement and bandwidth demand determining device
The particular procedure bandwidth of processing unit.It is to be herein pointed out in the device of the present invention, the number of plies of sampling filter layer, each layer
Down-sampled relation, the number of the dimensioning filtration module of the different bandwidth (yardstick) (being determined by resistance R0 sizes) in each layer
And the number of the gaussian filtering unit (resistor network) in dimensioning filtration module corresponding to each pixel signal is basis
Demand setting.
Following can performs step S540, constructs gaussian filtering process unit step, based on being constructed with active pull-up
The method of gaussian filtering network has spy according to the multiple specific active pull-ups of specific internetwork connection mode connection so as to form
Determine the gaussian filtering process unit of process bandwidth.
Step S550 is finally performed, constructing apparatus step, structure in S540 is connected based on the S520 network connection modes determined
The gaussian filtering process unit made is so as to forming described device.After device construction complete, light stream detector 101 is by the light stream of collection
After signal is converted to voltage signal, by the gaussian pyramid resistor network interconnecting relation of device 100, by corresponding voltage signal
The two-dimensional network being input in each layer, it is possible to realize different sizes, the gaussian filtering of different scale.
Next the specific effect based on a specific Simulation Application checking present invention.Circuit is emulated, Gao Sijin
Word tower completes the multi-scale filtering task to inputting initial data in analog domain, and processing time is the settling time of circuit.One
As for, if without electric capacity in circuit, circuit, which can be exceedingly fast, reaches stable, and its settling time can almost be ignored.But in practice
Parasitic capacitance, the layer capacitance of mos field effect transistor (MOS) etc. be present in circuit, therefore in simulation process
In need to take into account these factors.To different size of parasitic capacitance, its RC constant value is different, needed for circuit when establishing
Between it is also different, based on the actual estimated to circuit parasitic capacitance, the circuit simulated under several different size capacitive environments is built
Such as table 1 between immediately.
Parasitic capacitance (pico farad pf) | Circuit settling time (nanosecond ns) |
0.1 | 0.479 |
1 | 4.77 |
10 | 36.18 |
100 | 521.37 |
Table 1
The factor that influence to circuit energy consumption plays an important role is the size of filtering bandwidth, passes through experimental verification, filtering
Bandwidth and energy consumption are into inverse correlation, and bandwidth is bigger, and energy consumption is lower.By taking the image of 256x256 pixels as an example, if establishing 3 layers of golden word
Tower, every layer of pyramid have the two-dimentional resistor network of 5 different scales, and emulation counts the average energy consumption of each pixel and whole height
The relation of this pyramidal energy consumption and filtering bandwidth, if the pyramid bottom size of table 2 is 256 × 256, middle one layer is 128
× 128, the superiors are 64 × 64.
Filtering bandwidth | Power/pixel (milliwatt mW) | Gaussian pyramid total energy consumption (claps burnt pJ) |
4 | 0.0467 | 402.1 |
20 | 0.0140 | 120.3 |
40 | 0.0063 | 54.5 |
80 | 0.0042 | 36.3 |
100 | 0.0036 | 31.0 |
120 | 0.0030 | 25.4 |
Table 2
Using purely analog processing mode, device property and network topology structure is made full use of to complete nearly gaussian filtering effect
Fruit, in nanosecond rank the time required to whole processing procedure, exemplified by handling the image of 256 × 256 pixels, its power consumption is about tens
Burnt to hundreds of skins, energy consumption is extremely low.To sum up, compared with prior art, gaussian filtering process is carried out using the device of the present invention, not only
Processing speed is substantially increased, and reduces processing energy consumption.
While it is disclosed that embodiment as above, but described content only to facilitate understand the present invention and adopt
Embodiment, it is not limited to the present invention.Method of the present invention can also have other various embodiments.Without departing substantially from
In the case of essence of the present invention, those skilled in the art, which work as, can make various corresponding changes or become according to the present invention
Shape, but these corresponding changes or deformation should all belong to the scope of the claims of the present invention.
Claims (5)
1. a kind of multiple dimensioned gaussian filtering device for image analoging signal, it is characterised in that described device is used to treat place
Manage image analoging signal and carry out multiple dimensioned gaussian filtering so as to obtain filtered image analoging signal, wherein:
Described device includes multiple gaussian filtering units;
Negative resistance that each gaussian filtering unit is made up of multiple active circuits and common positive resistance are according to specific network
Connected mode connects and composes;
The pending image analoging signal includes multiple pixel analog signals, and the gaussian filtering unit is configured to one
Individual pixel analog signal carries out the gaussian filtering process under specific bandwidth to obtain filtered pixel analog signal;
Described device includes sampling filter layer, and the sampling filter layer is used to obtain the pending image based on particular sample rate
Corresponding multiple pixel analog signals and the progress Gauss of the multiple pixel analog signal to getting in analog signal
Filtering process;
The sampling filter layer includes multiple dimensioning filtration modules, and the dimensioning filtration module is used for specific bandwidth pair
Multiple pixel analog signals that the sampling filter layer is got carry out gaussian filtering process;
Each dimensioning filtration module includes multiple gaussian filtering units with same band, is filtered with some scale
In multiple pixel analog signals that sampling filter layer described in each gaussian filtering unit alignment processing in ripple module is got
A pixel analog signal, in any one dimensioning filtration module of the sampling filter layer, the gaussian filtering
Unit corresponds with each pixel analog signal that the sampling filter layer is got.
2. device as claimed in claim 1, it is characterised in that described device includes multiple sampling filter layers, each described to adopt
The corresponding specific sample rate of sample wave filtering layer, the different sampling filter layers correspond to different sample rates.
3. device as claimed in claim 1, it is characterised in that described device includes a sampling filter layer and multiple samplings
Extraction module, the sampling extraction module are used to from the filtered pixel analog signal of sampling filter layer output sieve
Analog signal corresponding to specific pixel is selected so as to obtain the filtered image analoging signal under corresponding sample rate.
4. such as the device any one of claim 1-3, it is characterised in that the sampling filter layer is filtered comprising multiple pixels
Ripple module, each pixel mould that each pixel filter module in the sampling filter layer is got with the sampling filter layer
Intend signal to correspond.
5. device as claimed in claim 4, it is characterised in that the pixel filter module includes multiple gaussian filtering lists
Member, the different gaussian filtering units in same pixel filter module have a different bandwidth, all in same sampling filter layer
Pixel filter module has identical structure.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510138175.3A CN104700373B (en) | 2015-03-26 | 2015-03-26 | A kind of multiple dimensioned gaussian filtering device and its building method for image analoging signal |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510138175.3A CN104700373B (en) | 2015-03-26 | 2015-03-26 | A kind of multiple dimensioned gaussian filtering device and its building method for image analoging signal |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104700373A CN104700373A (en) | 2015-06-10 |
CN104700373B true CN104700373B (en) | 2018-01-16 |
Family
ID=53347461
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510138175.3A Active CN104700373B (en) | 2015-03-26 | 2015-03-26 | A kind of multiple dimensioned gaussian filtering device and its building method for image analoging signal |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104700373B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105979174B (en) * | 2016-05-05 | 2018-11-23 | 清华大学 | A kind of filter network system and image processing system |
CN106097265A (en) * | 2016-06-08 | 2016-11-09 | 清华大学 | A kind of gaussian filtering device |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101714254A (en) * | 2009-11-16 | 2010-05-26 | 哈尔滨工业大学 | Registering control point extracting method combining multi-scale SIFT and area invariant moment features |
CN102324936A (en) * | 2011-06-30 | 2012-01-18 | 复旦大学 | Unbalanced spatial filter for edge maladjustment correction of resistance averaging network |
CN103234976A (en) * | 2013-04-03 | 2013-08-07 | 江南大学 | Warp knitting machine cloth flaw on-line visual inspection method based on Gabor transformation |
CN104408711A (en) * | 2014-10-30 | 2015-03-11 | 西北工业大学 | Multi-scale region fusion-based salient region detection method |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9014508B2 (en) * | 2012-04-24 | 2015-04-21 | Stmicroelectronics S.R.L. | Multiplierless coprocessor for difference of Gaussian (DoG) calculation |
-
2015
- 2015-03-26 CN CN201510138175.3A patent/CN104700373B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101714254A (en) * | 2009-11-16 | 2010-05-26 | 哈尔滨工业大学 | Registering control point extracting method combining multi-scale SIFT and area invariant moment features |
CN102324936A (en) * | 2011-06-30 | 2012-01-18 | 复旦大学 | Unbalanced spatial filter for edge maladjustment correction of resistance averaging network |
CN103234976A (en) * | 2013-04-03 | 2013-08-07 | 江南大学 | Warp knitting machine cloth flaw on-line visual inspection method based on Gabor transformation |
CN104408711A (en) * | 2014-10-30 | 2015-03-11 | 西北工业大学 | Multi-scale region fusion-based salient region detection method |
Non-Patent Citations (1)
Title |
---|
Physical Computing With No Clock to Implement the Gaussian Pyramid of SIFT Algorithm;Yi Li et al;《http://arxiv.org/》;20141231;第1-6页 * |
Also Published As
Publication number | Publication date |
---|---|
CN104700373A (en) | 2015-06-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105528638B (en) | The method that gray relative analysis method determines convolutional neural networks hidden layer characteristic pattern number | |
CN102243711B (en) | Neighbor embedding-based image super-resolution reconstruction method | |
DE112016005746T5 (en) | Folding neural network | |
CN106778745A (en) | A kind of licence plate recognition method and device, user equipment | |
CN105046672A (en) | Method for image super-resolution reconstruction | |
CN111950408B (en) | Finger vein image recognition method and device based on rule diagram and storage medium | |
CN104881682B (en) | A kind of image classification method based on guarantor office mapping and principal component analysis | |
CN107992891A (en) | Based on spectrum vector analysis multi-spectral remote sensing image change detecting method | |
CN113065526B (en) | Electroencephalogram signal classification method based on improved depth residual error grouping convolution network | |
CN106096660B (en) | Convolutional neural networks based on independent composition analysis algorithm | |
CN104834909B (en) | A kind of new image representation method based on Gabor comprehensive characteristics | |
CN108042132A (en) | Brain electrical feature extracting method based on DWT and EMD fusions CSP | |
CN108509991A (en) | Liver's pathological image sorting technique based on convolutional neural networks | |
CN109635811A (en) | The image analysis method of spatial plant | |
CN112732092B (en) | Surface electromyogram signal identification method based on double-view multi-scale convolution neural network | |
CN104700373B (en) | A kind of multiple dimensioned gaussian filtering device and its building method for image analoging signal | |
CN109886391A (en) | A kind of neural network compression method based on the positive and negative diagonal convolution in space | |
CN107661099A (en) | A kind of brain electricity analytical method for detecting the horizontal lower motion state of different Mental Workloads | |
CN103267941A (en) | Method for testing fault modes of integrated switching current circuit | |
CN109671055A (en) | Pulmonary nodule detection method and device | |
CN115100720A (en) | Low-resolution face recognition method | |
CN108710950A (en) | A kind of image quantization analysis method | |
CN110070486A (en) | A kind of image processing method, device and electronic equipment | |
CN115919330A (en) | EEG Emotional State Classification Method Based on Multi-level SE Attention and Graph Convolution | |
CN109784233B (en) | Phase-locked value weighted spatial filtering method and device based on Lp-norm |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |